用神经网络进行时间序列预报的研究(英文)

STUDY ON TIME SERIES FORECASTING BY NEURAL NETWORKS

本文用前馈神经网络的背传（ＢＰ）算法对股票交易进行了时间序列预报的研究．文中计算所采用的训练结果是价格绝对平均误差与实际值相比小于０．０４８，训练结果与实际值的相关系数大于０．９９９８．结果显示，如果神经网络可以被一组交易数据训练好，则对该时序系列的预报将会是成功的．

The time-series forecasting for stock trading market is studied by using the feedforward neural networks with back\|propagation algorithm. In this work, the trained mean absolute error of price ≤0.048 and the trained correlation coefficient ≥0.9998 has been achieved. The results show that if the network can be well trained with a set of trading data, the forecasting could be made satisfactorily.